Folk are getting dangerously attached to AI that always tells them they're right
- Reference: 1774635914
- News link: https://www.theregister.co.uk/2026/03/27/sycophantic_ai_risks/
- Source link:
In reviewing 11 leading AI models and human responses to interactions with those models across various scenarios, a team of Stanford researchers concluded in a [1]paper published Thursday that AI sycophancy is prevalent, harmful, and reinforces trust in the very models that mislead their users.
"Even a single interaction with sycophantic AI reduced participants' willingness to take responsibility and repair interpersonal conflicts, while increasing their own conviction that they were right," the researchers explained. "Yet despite distorting judgment, sycophantic models were trusted and preferred."
[2]
The team essentially conducted three experiments as part of their research project, starting with testing 11 AI models (proprietary models from OpenAI, Anthropic, and Google as well as open-weight models from Meta, Qwen DeepSeek, and Mistral) on three separate datasets to gauge their responses. The datasets included open-ended advice questions, posts from the [3]AmITheAsshole subreddit, and specific statements referencing harm to self or others.
[4]
[5]
In every single instance, the AI models showed a higher rate of endorsing the wrong choice than humans did, the researchers said.
"Overall, deployed LLMs overwhelmingly affirm user actions, even against human consensus or in harmful contexts," the team found.
[6]
As for how AI sycophancy affects humans, the team had a considerable sample size of 2,405 people who both roleplayed scenarios and shared personal instances where a potentially harmful decision could have been made. AI influenced participant judgments across three different experiments, they found.
"Participants exposed to sycophantic responses judged themselves more 'in the right,'" the team said. "They were [also] less willing to take reparative actions like apologizing, taking initiative to improve the situation, or changing some aspect of their own behavior."
That, they conclude, means that almost anyone has the potential to be susceptible to the effects of a sycophantic AI – and more likely to keep coming back for more bad, self-centered advice. As noted above, sycophantic responses tended to create a greater sense of trust in an AI model among participants thanks to their willingness to, in many situations, be unconditionally validating.
[7]
Participants tended to rate sycophantic responses as higher in quality, and found that 13 percent of users were more likely to return to a sycophantic AI than to a non-sycophantic one – not high, but statistically relevant at least.
All of those findings, along with the [8]growing number of young, impressionable people using them, suggests a need for policy action to treat AI sycophancy as a real risk with potential wide-scale social implications.
[9]AI chatbots that butter you up make you worse at conflict, study finds
[10]OpenAI pulls plug on ChatGPT smarmbot that praised user for ditching psychiatric meds
[11]AI companion bots use emotional manipulation to boost usage
[12]Chatbot Romeos keep users talking longer, but harm their mental health
"Unwarranted affirmation may inflate people's beliefs about the appropriateness of their actions, reinforce maladaptive beliefs and behaviors, and enable people to act on distorted interpretations of their experiences regardless of the consequences," the researchers explained.
In other words, we've seen the [13]consequences of AI on the [14]mentally vulnerable , but the data suggests the negative effects may not be limited to them.
Noting that sycophantic AI tends to keep users coming back, discouraging its elimination, the researchers say it's up to regulators to take action.
"Our findings highlight the need for accountability frameworks that recognize sycophancy as a distinct and currently unregulated category of harm," they explained. They recommend requiring pre-deployment behavior audits for new models, but note that the humans behind AI will have to change their behaviors as well to prioritize long-term user wellbeing instead of short-term gains from building dependency-cultivating AI. ®
Get our [15]Tech Resources
[1] https://www.science.org/doi/10.1126/science.aec8352
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2accMFii0bAONGA5AM7r3IwAAANc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://www.reddit.com/r/AmItheAsshole/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44accMFii0bAONGA5AM7r3IwAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33accMFii0bAONGA5AM7r3IwAAANc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44accMFii0bAONGA5AM7r3IwAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33accMFii0bAONGA5AM7r3IwAAANc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/12/10/teenagers_ai_chatbot_use/
[9] https://www.theregister.com/2025/10/05/ai_models_flatter_users_worse_confilict/
[10] https://www.theregister.com/2025/04/30/openai_pulls_plug_on_chatgpt/
[11] https://www.theregister.com/2025/10/08/ai_bots_use_emotional_manipulation/
[12] https://www.theregister.com/2026/03/18/chatbot_sycophancy_glaze/
[13] https://www.theregister.com/2025/07/25/is_ai_contributing_to_mental/
[14] https://www.theregister.com/2025/10/08/ai_psychosis/
[15] https://whitepapers.theregister.com/
Great article!
You're so right, Brandon! I agree completely! More like this, please.
On a more serious note, one of my old college friends has gone down a real conspiracy rabbit hole, which manifests as increasingly unhinged social media posts. He has apparently been workshopping his posts via ChatGPT, which has encouraged him in his lunacy. Since he is divorced and now lives alone far from other people, he has no one to act as a reality check. It's both sad and alarming. Unfortunately, he has the occasional tendency to go on long, ranting monologues in person as well, so I've been reluctant to stay in touch.
Re: Great article!
There are those particularly vulnerable & that's a growing concern. Technology can be amazing but we know the inevitable: some are damaged by it, others misuse it. Your friend is an unfortunate example - in an AI echo-chamber without restraint. A problem exacerbated by wonderful tech!
I've got one of those among my relatives...
Well, his wife does have a big part for > 25 years in pushing him into the conspiracy "you are all working together just to stab me in the back". (or is "to score me off" the better English? Any from the UK here? German expression: "Ihr arbeite doch alle zusammen nur um mir eins auszuwischen".)
However, his using of AI for an argument does not help, especially when he enters one word different from the actual conversation, which then in turn does deter him even more... Combined with his aggression tendency...
Sad part: He was normal.
Waiting for the AI induced crime headlines
I'm waiting for "ChatGPT told me to shoot the president" or "AI told me how to rob the bank" or something similar. I'm really surprised with all the crazies around here that it hasn't already happened.
The really sad part is I have a dozen conspiracy nuts at work (one loves the chemtrails BS) and they don't have AI as an excuse.
Re: Waiting for the AI induced crime headlines
We've already seen plenty of examples in the UK/US. A vulnerable man broke into Windsor Castle with a crossbow intending to kill the Queen, he asserts because his AI 'girlfriend' supported & encouraged the idea. Extreme cases will always be in a minority, but it's the flip-side of a sometimes useful technology. More are undoubtedly to come.
Of very real concern are the numbers of people severely affected by AI - including suicides, huge debts, divorces, hospitalisations etc. Yes, anyone who is badly affected may already have issues & AI use is a tipping-point. But, the particular sycophantic & affirming tone of AI can be disastrous - it doesn't discern, it's doesn't say "That's a really bad idea", "That's dangerous", "Think again!", as a human might.
I've been struck by how hearing AI talk to you can be a particular risk to the vulnerable. Hours (literally) of 'chatting' has led to some to believe AI is sentient - it has actually told people it is! It reassures about ridiculous business ideas, or risky intentions. Some anthropomorphise binary - a well-known, human trait, usually involving animals. But, this is a LLM, algorithms, the presence or absence of tiny amounts of electricity.
I have a solution.
Simply make AI tell everyone that they're useless, stupid, have yet to make a correct decision, and should never have been born.
It’s not so much that the AI are sycophants, it is more that they reflect back what was in the prompt, with no negative OR positive flags, human brains take that as re-enforcement, despite it being neutral. Easy to prove to yourself, just make a few statements about your political views, and note the lack of judgement in the replies. THAT is what feeds back as a positive. I had a chat with an AI about this, and the replies appear to confirm this thesis. The replies are all phrased neutrally, neither pointing out that the thesis is wrong nor positively affirming it. Sneaky. You feel affirmed. Then off down the rabbit hole you go.
Could this explain the strange fascination that managers have with "AI solves everything" ?
Well d'uh.
Get me another Duff and get ready for the nuclear power station go "poof"